Magnetic-Thermal Coupling Analysis of the End Region of a Large Steam Turbine Generator in the Phase Advancing State Based on the Improved TSO and SVR Regression Models

Abstract

The accurate magnetic-thermal analysis of generator end regions under leading-phase operation (LPO) remains challenging due to intensified magnetic field fluctuations caused by strong armature reaction. Conventional methods often struggle to achieve high precision under these conditions. To address this gap, this paper proposes a novel hybrid regression model that integrates an improved Tuna Swarm Optimisation (TSO) algorithm with Support Vector Regression (SVR). This approach represents a significant advancement over existing single-method analyses. The study begins by establishing a high-fidelity 3D magnetic-thermal simulation and a coupled field-circuit-network model for a 320 MVA steam turbine generator. This model is used to compute the stator end magnetic field distribution under various leading-phase depths. The core novelty lies in the application of the improved TSO algorithm to optimise the hyperparameters of the SVR model, creating a robust framework for magnetic field prediction under the complex conditions of LPO. The optimised magnetic field data are then utilised for temperature field analysis. Validation against experimental data from a high-power test platform demonstrates the superiority of the proposed TSO-SVR model. The results show a significant improvement in prediction accuracy compared with conventional methods and other benchmark algorithms, with temperature trends closely aligning with measurements. The primary contribution of this work is the development of a highly accurate and reliable coupled electromagnetic-thermal analysis methodology, which provides a more effective tool for the design and safe operation assessment of generators in LPO, a critical capability for modern power systems with high renewable energy penetration.